Multi-level Local Binary Pattern Analysis for Texture Characterization

Texture is the core element in numerous computer vision applications. The objective of this paper is to present a novel methodology for learning and recognizing textures. A local binary pattern (LBP) operator offers an efficient way of analyzing textures. A multi-level local binary pattern operator which is an extension of LBP is proposed for extracting texture feature from the images. The operator finds the association of LBP operators at multiple levels. This association helps to identify macro features. Depending on the size of the operator, octets are framed and LBP responses of the octets are noted. Their occurrence histograms are combined to frame the texture descriptor. The proposed operator is gray-scale invariant. The operator is computationally simple since it can be realized with few operations in the local neighborhood. A non-parametric statistic named G-Statistic is used in the classification phase. The classifier is trained with images of known texture class to build a model for that class. Experimental results prove that the approach provides good discrimination between the textures.